AI Compute Contracts Top $2.3 Trillion, Payment Reset Wall Looms for 2027-2028
Analysis finds over $2.3 trillion in AI compute contracts have been signed, with payments set to surge in 2027-2028, resembling the subprime reset wall of 2007-2008. OpenAI and Anthropic account for about $712 billion in contract value, while the four largest US cloud providers hold roughly $2.3 trillion in backlog. Unrecorded take-or-pay obligations, estimated at $1.66 trillion in present value, add to $470 billion in reported debt, totaling about $2.1 trillion in economic commitments.
The artificial intelligence boom is approaching a structural turning point. By August 2026, investment in AI infrastructure had reached unprecedented levels, but the financing architecture underpinning this surge is showing characteristics similar to the 2008 subprime mortgage crisis. More than $2.3 trillion in compute contracts have been signed and recorded in the remaining performance obligations and contract backlogs of the four largest US cloud service providers, yet most of these contracts have not yet begun generating bills. At the core of these agreements are take-or-pay compute contracts, under which payments do not commence at signing but only after data centers are built and capacity is delivered, with construction typically taking 24 to 36 months.
This structure bears resemblance to adjustable-rate mortgages in the earlier property cycle. Subprime mortgages issued in 2005-2006 underwent a concentrated reset after a two-year teaser period, forming the 2007-2008 reset wall. The current peak in compute contract signings occurred in 2025-2026, and based on construction timelines, a concentrated activation period is expected in 2027-2028. Disclosures indicate that OpenAI and Anthropic together have approximately $712 billion in nominal contract amounts set to convert to full payments in 2027-2028. As of the second quarter of 2026, the four largest US cloud providers held about $2.3 trillion in contracted revenue backlog, of which roughly $1 trillion came from OpenAI and Anthropic.
Take-or-pay contracts carry debt-like characteristics in economic substance. Rating agencies have for over three decades treated such obligations as deemed debt, but this practice has not yet been applied to compute contracts. Estimates suggest that AI-related companies report total debt of about $470 billion, while the present value of disclosed non-cancellable compute and capacity commitments is approximately $1.66 trillion, bringing total economic obligations to about $2.1 trillion. For comparison, in March 2007, US subprime mortgage balances stood at approximately $1.3 trillion. These contractual obligations are not recorded on balance sheets, primarily due to disclosure asymmetry, the fact that the largest buyers are private companies, and the structuring of contracts as service agreements rather than leases.
OpenAI signed approximately $1.2 trillion in compute commitments between June and December 2025, with deals announced within about three weeks in October 2025 having a combined nominal value exceeding the market capitalization of 95% of companies in the S&P 500. The company's unaudited financial statements as of March 31, 2026, disclosed $665 billion in non-cancellable compute commitments, with management's latest planned amount at $750 billion. Under various revenue growth scenarios, even based on management's own plan, compute costs as a percentage of revenue would exceed 200% at the 2027 peak. The company is valued at over $850 billion, but if take-or-pay obligations are counted as senior debt at present value, they would amount to approximately $450-500 billion, exceeding half of its equity value.
Compute contracts signed in 2025-2026 will see concentrated activation in 2027-2028, when payment obligations become enforceable under contract terms, regardless of actual utilization and revenue levels.